scripts/run_train_v25.sh

#!/usr/bin/env bash
# v25 training — T40 fix: address 3 remaining Track 4 failures from v24.
#
# Changes from v24:
#   1. DEGRADATION_CONCEPT_MAP (T40 in src/losses/degradation_ranking.py):
#      blur:  [1, 2]    → [1, 2, 0]  (+orient_coh: blur smears orientation fields)
#      noise: [3]       → [3, 4]     (+contrast_u: noise disrupts local contrast)
#      jpeg:  [2, 1]    → [2, 1, 4]  (+contrast_u: JPEG blocking creates contrast bands)
#      dry_skin/wet_press/occlusion: unchanged from T39
#
#   2. --concept-deg-gamma 1.5  (down from 2.0)
#      gamma=2.0 caused noise→noise_lv REGRESSION (+0.365 in v24).
#      gamma=1.5 keeps enough signal for blur/jpeg (needed >0.5) while
#      reducing saturation pressure that triggered noise_lv inversion.
#
# Expected improvements:
#   noise → noise_lv:     +0.365 → negative  (gamma reduced + noise now 2-concept)
#   dry_skin → contrast_u: +0.051 → negative  (3 degradations now target contrast_u)
#   dry_skin → orient_coh: +0.008 → negative  (blur gives orient_coh strong signal)

set -euo pipefail

REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
export PATH="/home/aiserver/miniconda3/bin:$PATH"

VERSION="v25"
SAVE_DIR="${REPO_ROOT}/sifq/checkpoints/${VERSION}"
LOG_FILE="${REPO_ROOT}/sifq/logs/train_${VERSION}.log"
EVAL_SCRIPT="${REPO_ROOT}/sifq/scripts/run_eval_${VERSION}.sh"

mkdir -p "${REPO_ROOT}/sifq/logs"

python "${REPO_ROOT}/sifq/scripts/train_sifq.py" \
  --root-302a  "${REPO_ROOT}/dataset/302a/images/challengers" \
  --root-302b  "${REPO_ROOT}/dataset/302b/images/baseline" \
  --root-302d  "${REPO_ROOT}/dataset/nist_302d/images/auxiliary" \
  --root-fvc2002 "${REPO_ROOT}/dataset/FVC_Dataset/FVC2002" \
  --root-fvc2004 "${REPO_ROOT}/dataset/FVC_Dataset/FVC2004" \
  --root-polyu "${REPO_ROOT}/dataset/PolyU" \
  --exclude-sensor "R_1000_slap,R_500_slap,S_500_slap" \
  --mdgt-checkpoint "${REPO_ROOT}/pad/TRAM-downstream/checkpoint/checkpoints_dinov2_tram/best_eer.pt" \
  --epochs 60 \
  --batch-size 96 \
  --image-size 224 \
  --lr 1e-4 \
  --spread-mode uniform \
  --spread-weight 3.0 \
  --concept-deg-gamma 1.5 \
  --sd302-concept-weight 0.0 \
  --deg-every-n-steps 2 \
  --no-mat-stats \
  --proto-max-batches 0 \
  --k-cross 0 \
  --max-train-samples -1 \
  --num-workers 8 \
  --gpus 0 \
  --save-dir "${SAVE_DIR}"

echo "[auto-eval] Training done. Starting eval ${VERSION}..."
bash "${EVAL_SCRIPT}" >> "${LOG_FILE}" 2>&1